The graduate and advanced undergraduate students in Nicola Bezzo's Autonomous Mobile Robots course, along with Bezzo's colleague Roman Krzysztofowicz in the bowtie, gather to find out which team’s TurtleBot will take the prize. (Photos by Matt Cosner)
In the tale of the tortoise and the hare, slow and steady won the race.
In the competition among robots working their way through a carboard maze last week at the University of Virginia School of Engineering and Applied Science, speed and precision formed the winning combination.
Some of these are people who never programmed a robot before, so it's quite impressive.
Nicola Bezzo
The maze runners were part of associate professor Nicola Bezzo’s annual Autonomous Mobile Robots competition, from his course of the same name.
The speeds clocked by the low-cost, open-source learning platform were indeed befitting of the name “TurtleBot.” But being methodical was just as important as being relatively fast. Low enough to the ground to invoke the image of a tortoise in the mind, the stripped-down black robots could very well be R2D2’s undercarriage. Their bases run on the same technology as the roaming Roomba home vacuums, Bezzo noted. LiDar, the spinning, hockey puck shaped disc at the front of the TurtleBots shoots an invisible laser ahead of its path, potentially helping the robots detect walls and differentiate incidental gaps from actual pathways. Students wrote code that responded to the sensor inputs, so that the robots could "learn" as they moved.
Team No. 1 unpacks their TurtleBot. They enter the competition with about "75% confidence," they say.
Though TurtleBot has camera capability, students are only allowed to use the LiDar (laser sensing).
Bezzo sets the course, which is new to the students and their robots for the competition. They trained on a different configuration.
The professor explains the rules: Each team must run two official time trials within five minutes. The team with the least points wins.
Teams persevere through glitches and use some of the "extra" time within their five minutes to perform test runs.
Some students naturally document their performance on their smartphones.
This TurtleBot gets a touch too close to a wall. Penalty points are added.
Team No. 7 strategizes between trials. They successfully navigate two runs, clearing all check points and avoiding walls.
Team No. 5 confers during testing to see if they can top Team No. 7's dauntingly low score.
And the Winners Are...
Having run two clear paths of 21.6 seconds each, no penalties, the winning group was Benjamin Dacey, Dylan Do, Benjamin Pusch and Charles Wilmot of Team No. 7.
Teams No. 1 and No. 5 were the runners-up.
Some teams chose to use the gap detection code the professor provided. Some chose to map the maze at run time. Others chose to create a wall-following behavior or to make the robot stop, scan and then pick the largest open space. What was the winning secret?
For this maze, it was keeping things simple. “It was just pure obstacle avoidance,” Wilmot said. “We kept the robot going straight until it detected an obstacle. We had four sections of our LiDar. We summed them up and then pointed.”
Pusch added, “We just thought, what’s the simplest way to avoid one obstacle and then it turned out to work for the maze [in testing configuration]. We tried to employ these other mappings and nothing ended up being faster.”
Dylan Do, Benjamin Dacey, Charles Wilmot and Benjamin Pusch had all their ducks in a row.
